# US 2013/0158996 A1 · Acoustic Processing Unit

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2013/0158996 A1
- Document ID: US20130158996A1
- Status: Published application
- Kind: A1
- Application number: 13/489,799
- Filed: 2012-06-06
- Published: 2013-06-20
- Earliest claimed: 2011-12-19
- Assignees: Spansion LLC
- Inventors: Richard Fastow; Jens Olson
- Web page: https://patentgrep.com/patent/US20130158996A1

## CPC

- G06F 17/10: Complex mathematical operations {(function generation by table look-up G06F1/03; evaluation of elementary functions by calculation G06F7/544)}
- G10L 15/08: Speech classification or search
- G10L 15/28: Constructional details of speech recognition systems
- G10L 15/02: Feature extraction for speech recognition; Selection of recognition unit
- G10L 15/14: using statistical models, e.g. Hidden Markov Models [HMMs] (G10L15/18 takes precedence)
- G10L 15/187: Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

Embodiments of the present invention include an apparatus, method, and system for acoustic modeling. The apparatus can include a senone scoring unit (SSU) control module, a distance calculator, and an addition module. The SSU control module can be configured to receive a feature vector. The distance calculator can be configured to receive a plurality of Gaussian probability distributions via a data bus having a width of at least one Gaussian probability distribution and the feature vector from the SSU control module. The distance calculator can include a plurality of arithmetic logic units to calculate a plurality of dimension distance scores and an accumulator to sum the dimension distance scores to generate a Gaussian distance score. Further, the addition module is configured to sum a plurality of Gaussian distance scores to generate a senone score.

## Claims

### Claim 1

A senone scoring unit (SSU) comprising:
an SSU control module configured to receive a feature vector;
a distance calculator configured to receive a plurality of Gaussian probability distributions via a data bus having a width of at least one Gaussian probability distribution and the feature vector from the SSU control module, the distance calculator comprising:
a plurality of arithmetic logic units (ALUs), wherein each of the ALUs is configured to receive a portion of the at least one Gaussian probability distribution and to calculate a dimension distance score between a dimension of the feature vector and a corresponding dimension of the at least one Gaussian probability distribution; and
an accumulator configured to sum the dimension distance scores from the plurality of ALUs to generate a Gaussian distance score; and
an addition module configured to sum a plurality of Gaussian distance scores corresponding to the plurality of Gaussian probability distributions to generate a senone score.

### Claim 2 (depends on 1)

The senone scoring unit of claim 1, further comprising:
a feature vector matrix module configured to store a scaling factor for the dimension of the feature vector.

### Claim 3 (depends on 1)

The senone scoring unit of claim 1, wherein the SSU control module comprises an input buffer configured to store an active senone list, wherein the active senone list comprises a base address associated with an address space of a memory device and one or more indices related to the base address at which one or more senones are located in the memory device.

### Claim 4 (depends on 1)

The senone scoring unit of claim 1, wherein the SSU control module is configured to generate an SSU clock signal for the distance calculator and the addition module, wherein a frequency of the SSU clock signal is a clock-divided representation of a frequency at which the at least one Gaussian probability distribution is read from a memory device.

### Claim 5 (depends on 1)

The senone scoring unit of claim 1, wherein the Gaussian probability distribution is 768 bits and the width of the data bus is 768 bits.

### Claim 6 (depends on 1)

The senone scoring unit of claim 1, wherein the distance calculator further comprises:
a datapath multiplexer configured to distribute information associated with the dimension of the feature vector, a mean value associated with the corresponding dimension of the at least one Gaussian probability distribution, and a variance value associated with the corresponding dimension of the at least one Gaussian probability distribution to the plurality of ALUs.

### Claim 7 (depends on 1)

The senone scoring unit of claim 1, wherein the plurality of ALUs and the accumulator are configured to generate the Gaussian distance score over a predetermined number of SSU clock cycles, wherein the predetermined number of SSU clock cycles equates to a read access time of the at least one Gaussian probability distribution from a memory device.

### Claim 8

A method for acoustic modeling, the method comprising:
receiving a plurality of Gaussian probability distributions via a data bus having a width of at least one Gaussian probability distribution and a feature vector from an external computing device;
calculating a plurality of dimension distance scores based on a plurality of dimensions of the feature vector and a corresponding plurality of dimensions of the at least one Gaussian probability distribution;
summing the plurality of dimension distance scores to generate a Gaussian distance score for the at least one Gaussian probability distribution; and
summing a plurality of Gaussian distance scores corresponding to the plurality of Gaussian probability distributions to generate a senone score.

### Claim 9 (depends on 8)

The method of claim 8, wherein the receiving comprises distributing information associated with the plurality of dimensions of the feature vector, a plurality of mean values associated with the corresponding plurality of dimensions of the at least one Gaussian probability distribution, and a plurality of variance values associated with the corresponding plurality of dimensions of the at least one Gaussian probability distribution.

### Claim 10 (depends on 8)

The method of claim 8, wherein the calculating comprises:
storing a plurality of scaling factors corresponding to the plurality of dimensions of the feature vector; and
applying the plurality of scaling factors to the plurality of dimensions of the feature vector during the calculation of the plurality of dimension distance scores.

### Claim 11 (depends on 8)

The method of claim 8, wherein the calculating comprises calculating the plurality of dimension distance scores based on at least one senone from an active senone list, wherein the active senone list comprises a base address associated with an address space of a memory device and one or more indices related to the base address at which the at least one senone is located in the memory device.

### Claim 12 (depends on 8)

The method of claim 8, wherein the summing the plurality of dimension distance scores comprises generating the Gaussian distance score over a predetermined number of senone scoring unit (SSU) clock cycles, wherein the predetermined number of SSU clock cycles equates to a read access time of the at least one Gaussian probability distribution from a memory device.

### Claim 13

A system for acoustic modeling, the system comprising:
a memory module configured to interface with an external computing device to receive a feature vector; and
a senone scoring unit (SSU) comprising:
a distance calculator configured to receive a plurality of Gaussian probability distributions from the memory module via a data bus having a width of at least one Gaussian probability distribution and the feature vector from the memory module, the distance calculator comprising:
a plurality of arithmetic logic units (ALUs), wherein each of the ALUs is configured to receive a portion of the at least one Gaussian probability distribution and to calculate a dimension distance score between a dimension of the feature vector and a corresponding dimension of the at least one Gaussian probability distribution; and
an accumulator configured to sum the dimension distance scores from the plurality of ALUs to generate a Gaussian distance score; and
an addition module configured to sum a plurality of Gaussian distance scores corresponding to the plurality of Gaussian probability distributions to generate a senone score.

### Claim 14 (depends on 13)

The system of claim 13, wherein the memory module and SSU are integrated on the same chip.

### Claim 15 (depends on 13)

The system of claim 13, wherein the memory module are SSU are configured to operate in two different clock domains, the SSU operating at a SSU clock frequency that is a clock-divided representation of a memory module clock frequency.

### Claim 16 (depends on 13)

The system of claim 13, wherein the memory module comprises a memory device configured to store an acoustic library that includes the plurality of Gaussian probability distributions.

### Claim 17 (depends on 13)

The system of claim 13, wherein the SSU further comprises:
a feature vector matrix module configured to store a scaling factor for the dimension of the feature vector.

### Claim 18 (depends on 13)

The system of claim 13, wherein the SSU comprises an input buffer configured to store an active senone list, wherein the active senone list comprises a base address associated with an address space of the memory module and one or more indices related to the base address at which one or more senones are located in the memory module.

### Claim 19 (depends on 13)

The system of claim 13, wherein the SSU is configured to generate an SS clock signal for the distance calculator and the addition module, wherein a frequency of the SSU clock signal is a clock-divided representation of a frequency at which the at least one Gaussian probability distribution is read from the memory module.

### Claim 20 (depends on 13)

The system of claim 13, wherein the plurality of ALUs and accumulator are configured to generate the Gaussian distance score over a predetermined number of SSU clock cycles, wherein the predetermined number of SSU clock cycles equates to a read access time of the at least one Gaussian probability distribution from the memory module.
